Looking at 2025, Tech: Rising above the AI line

Takeover by AI agents

The adoption of AI agents that can perform certain tasks without any guidance, learn from their mistakes, and even make decisions on these tasks, will become the more popular way to adopt AI for enterprise as well as consumer uses.

It is easy to create AI agents for a variety of use cases, and to train them to take care of mundane processes that do not need human intervention, though alerting humans when something needs their attention. As of December 2024, users have created more than 3 million custom GPTs on ChatGPT, only about 5% of which are publicly accessible through the GPT Store.

Soon, a lot of work on the Internet, on apps, and software can be executed by these personalised agents, although we are still waiting for companies to release consumer-facing agents that solve specific problems that users might have. One impact of this will be a shift beyond the chat-driven paradigm that GenAI processes are now dependent on — these agents will be pre-programmed, and will not need prompts to get on with what they are supposed to do.

This will also raise some ethical questions — for instance when you get an agent to reply to emails sent by human contacts expecting you to respond. Banks could already be using specific AI agents to evaluate loan applications on specific criteria before they reach the desk of a manager, and hospitals could be using agents to highlight critical points in blood reports to doctors before they go on their rounds.

Death of dashboards

AI has started to make data more accessible, removing hurdles of skill behind analysing large data sets. This will gradually lead to dashboards being replaced by GenAI tools that answer specific data questions with visualisations, trend lines, or even predictions based on their visibility.

You could see products that show analytics move to this format by the end of 2025, even though the costs involved in handling these queries might limit the conversational data analytics to enterprise products for some time.

At the same time, data itself will have to move to formats that support the requirements of these new tools. As user queries will no longer be limited by the headers on a spreadsheet, data collection will need to become more complex and detailed in order to capture all the metrics that could possibly be summoned.

Rise of not-so-social media

Social media platforms are in the midst of a crisis. TikTok is staring at an uncertain future despite its massive popularity in the United States and other markets. X users unhappy with the way Elon Musk is using the platform are moving en masse to Bluesky. Facebook struggles to find younger users, and Threads any serious engagement. Instagram continues to grow, though — with some help from dodgy content and even more questionable content moderation.

All this could produce a whole new platform that would sweep up audiences — even though such attempts have not been successful in the past. Interestingly, a lot of localised platforms, even in India, have died down over the past year or so, having realised that engagement induced by Covid-19 pandemic lockdowns was not something they could sustain when users have other things to do.

Making matters worse, younger users have found ways to socialise on not-so-social platforms where closed groups make more sense than the bulletin board formats of the now archaic platforms of the desktop era.

Unmatched computing power

A lot is happening in the world of computing. Google Willow quantum chips have shown the potential to take users to a whole new level, and NVIDIA’s new GPUs can power AI to solve the biggest problems of our times much faster than we could ever imagine. All this is happening as the new Apple M4 processors bring professional-grade processing power to even portable PCs like the Mini.

With a new wave of devices expected soon, processing times could be cut like never before, opening up new possibilities on what computers can do. But even as computational power scales up, questions will be asked on whether the data models are good or complex enough for these computers to put to good use. The ‘data wall’ with maxed-out computing capabilities could start pushing AI models to their limits and potentially trigger a pivot to new trajectories of reasoning.

AI meets hardware

The first few AI-led hardware products such as the Humane AI Pin and Rabbit r1 have been disasters. This is where companies have realised again that being an early mover is not always an advantage, especially when the technology is nascent.

In the new year, expect a new wave of products that use AI at their core, and maybe solve more specific problems instead of trying to be a smartphone equivalent. It’s difficult to predict what these devices will be. Some of them could be unveiled at the CES in early January.

Expect some experimentation on getting AI platforms to take the lead on smartphones and laptops — how about an OpenAI phone full of agents that replace apps or a Perplexity laptop that replaces the OS with an AI chat?

The possibilities are endless — expect all possible paths to be trod.

Nandagopal Rajan is Chief Operating Officer, The Indian Express Digital

NEXT: DIPLOMACY